JP2000251015A - Method and device for discriminating similar character and storage medium recording similar character discrimination program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a similar character discriminating method capable of exactly discriminating similar characters while flexibly dealing with a document mixing various fonts or emphasized expressions. SOLUTION: Concerning a character recognizing device provided with a pattern recognizing means for recognizing respective characters described on an original on the basis of character patterns, it is discriminated whether the height of respective character patterns is uniform or not by receiving a character pattern stream corresponding to character strings within a prescribed range including similar characters and the recognized result of the pattern recognizing means, when there is a dispersion in the height of character patterns, the similar characters are discriminated by classification into a capital letter class and a small letter class with discrimination analysis concerning the height of character patterns and when the height of character patterns is uniform, the fixed characters included in the recognized result are classified into prescribed letter types. Then, similar characters are discriminated corresponding to the result of comparison between the height of the character pattern corresponding to the similar characters and the height of the character pattern classified into the prescribed letter type.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device provided in a document file system and the like, and more particularly to a technique for determining a similar character which is difficult to determine because it is represented by a similar character pattern. It is. For example,
Characters whose uppercase and lowercase characters are represented by character patterns that are almost similar, such as uppercase “C” and lowercase “c” and lowercase “tsu” and lowercase “tsu” (hereinafter referred to as “hi”). , And similar characters), the recognition result cannot be determined only by a pattern matching technique using a pattern dictionary or the like. For this reason, there is a need for a technique for distinguishing between uppercase and lowercase, in addition to character recognition by pattern matching.

[0002]

2. Description of the Related Art As a technique for determining whether a similar character is uppercase or lowercase, Japanese Unexamined Patent Publication No.
Japanese Patent Application Publication No. 11983, “Similar character identification method” has been proposed. FIG. 10 shows an example of the configuration of a character recognition device to which a similar character identification method according to the technique described in Japanese Patent Application Laid-Open No. 3-111983 is applied.

[0003] The pattern recognition unit 402 shown in FIG.
A configuration in which the document reading unit 401 cuts out a character pattern corresponding to each character from the image data read from the document, and recognizes each character by comparing the character pattern with a standard character pattern registered in the pattern dictionary 403. It has become. In FIG. 10, when the similar character extracting unit 404 obtains a recognition result indicating a similar character by the pattern recognizing unit 402, the similar character extracting unit 404 sends the corresponding character pattern to the similar character determining unit 410, and accordingly, The similar character determination unit 410 performs a determination process described later.

In FIG. 10, an output processing unit 405
Is configured to receive a determination result obtained by the similar character determination unit 410, include the determination result in the recognition result, and send the result to the document file system. In the similar character determination unit 410, the blank ratio calculation unit 411 calculates the ratio of the width of the blank portion of the character pattern representing the character to be determined to the standard line width received from the pattern recognition unit 402, and 412.

The comparator 412 compares the obtained blank ratio with a predetermined threshold value TH1, and outputs a determination result indicating that a lowercase letter is present when the blank ratio is equal to or greater than the threshold value TH1. In the case of the following, a configuration may be adopted in which a determination result indicating that the character is a capital letter is output. Here, as shown in FIG. 11, if the font and the number of points are the same, the upper space dl corresponding to uppercase letters and the upper space ds corresponding to lowercase letters are significantly different. It is possible to easily determine whether the character is uppercase or lowercase based on the ratio of the upper space d corresponding to the character to be determined.

[0006]

By the way, with the spread of word processors and the like, diversification of documents to be recognized by the character recognition device has progressed, and a variety of fonts have been switched, and a variety of fonts using emphasized expressions such as italic characters and bold characters have been used. Various expression techniques have come to be used. In such a document in which various fonts and emphasized expressions are mixed, when the ratio between the upper space dl corresponding to the character pattern representing an uppercase letter and the standard line width D exceeds the above-described predetermined threshold TH1, or conversely, It is conceivable that the ratio between the upper blank space ds corresponding to the character pattern representing lowercase letters and the standard line width D is lower than the above-mentioned predetermined threshold value TH1.

However, in the above-described conventional case determination method, since the determination is made based on the comparison result between the blank ratio of the character to be determined and the fixed threshold value TH1, similar characters included in such a document are used. Is difficult to determine accurately. The present invention provides a similar character discrimination method and a similar character discrimination device capable of flexibly coping with a document in which various fonts and emphasized expressions are mixed and accurately determining similar characters, and a storage medium storing a similar character discrimination program. The purpose is to do.

[0008]

FIG. 1 shows the principle of a similar character discriminating method according to the present invention. According to a first aspect of the present invention, there is provided a character recognition apparatus including a pattern recognition unit for recognizing each character written on a document based on a character pattern read from the document. A character pattern string corresponding to a predetermined range of character strings including similar characters to be recognized and a recognition result performed on the character pattern string by the pattern recognition unit are received, and the height of the character pattern included in the character pattern string is one. Is determined, and if the heights of the character patterns included in the character pattern string vary, the height equivalent to the capital letters is determined by discriminant analysis of the height of the character pattern included in the character pattern string. Class into uppercase classes with heights and lowercase classes with heights equivalent to lowercase,
The similar character recognition result is determined based on the similar character classification result, and if the height of the character pattern included in the character pattern string is uniform, the character is determined for the determined character included in the recognition result. Classification into multiple character types with different distribution characteristics in the height direction of the pattern, the height of character patterns corresponding to similar characters, and the height of character patterns corresponding to confirmed characters classified into multiple character types Is characterized in that a recognition result corresponding to a similar character is determined according to the comparison result.

According to the first aspect of the present invention, when there is a variation in the height of a character pattern representing a character string including similar characters, discriminant analysis is performed by focusing on the character height, and each character included in the character set is determined. By classifying into uppercase and lowercase classes, it is possible to determine whether similar characters to be determined are uppercase or lowercase using the heights of all characters included in the character set.

If the height of the character pattern is uniform, the height of a character belonging to a character type that is a reference for the height of uppercase letters or lowercase letters among characters belonging to a character set is determined. Utilization can be used to distinguish similar characters individually. FIG. 2 shows a principle block diagram of the similar character discriminating apparatus of the present invention. According to a second aspect of the present invention, there is provided a character recognition apparatus including a pattern recognition unit for recognizing each character written on a document based on a character pattern read from the document. A character set input unit 111 for receiving and inputting a character pattern string corresponding to a predetermined range of character strings including similar characters to be represented and a recognition result performed on the character pattern string by the pattern recognition unit 101; A height distribution determining unit 112 for determining whether the height of each character pattern included in the column is uniform, and a character pattern sequence according to a determination result that the height of each character pattern varies. Based on the discriminant analysis of the height of the character pattern included in, the uppercase class with the height equivalent to uppercase and the height equivalent to lowercase A classification analysis processing unit 113 that classifies the character patterns into character classes and determines a recognition result of similar characters to be determined based on the classification result; and a determination unit that determines that the height of each character pattern is uniform. hand,
A classification unit 114 for classifying the determined characters included in the recognition result into a plurality of character types having different distribution characteristics in the height direction of the character pattern; a height of the character pattern corresponding to the similar character; Based on the height of the character pattern corresponding to the confirmed character classified into
And an individual determination unit for individually determining a recognition result corresponding to each similar character.

According to a second aspect of the present invention, a height distribution determining means is provided.
2 operates the discriminant analysis processing means 113 or the classification means 114 and the individual judgment means 115 in response to the height of a series of character patterns received from the pattern recognition means 101 via the character set input means 111. Utilizing the characteristics of the distribution, similar characters included in this character set can be determined.

According to a third aspect of the present invention, there is provided a character recognition apparatus provided with a pattern recognition means for recognizing each character written on a document based on a character pattern read from the document. A character set inputting step of receiving and inputting a character pattern string corresponding to a predetermined range of character strings including similar characters represented by lowercase letters and a recognition result performed on the character pattern string by the pattern recognition means 101; A height distribution determination procedure for determining whether or not the height of each character pattern included in the character pattern string is uniform, and a character pattern according to a determination result that the height of each character pattern varies. By discriminant analysis on the height of the character pattern included in the column,
Classification into uppercase class with height equivalent to uppercase and lowercase class with height equivalent to lowercase, and based on this classification result, a discriminant analysis processing procedure to determine the recognition result of similar characters to be discriminated. According to the determination result that the height of each character pattern is uniform, the determined characters included in the recognition result are classified into a plurality of character types having different distribution characteristics in the height direction of the character pattern. Based on the classification procedure, the height of the character pattern corresponding to the similar character, and the height of the character pattern corresponding to the determined character classified into a plurality of character types, individually recognize the recognition result corresponding to each similar character. The individual determination procedure to be determined is executed by a computer.

According to a third aspect of the present invention, a discriminant analysis processing procedure, a classification procedure, and an individual determination procedure are executed in accordance with a determination result in the height distribution determination procedure, so that a pattern is received from the pattern recognition means 101 by a character set input procedure. Utilizing the height distribution characteristics of a series of character patterns
Similar characters included in this character set can be determined.

[0014]

Embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 3 shows a configuration of a character recognition device to which the similar character discrimination device of the present invention is applied.

In the character recognition apparatus shown in FIG. 3, a pattern recognition unit 402 sequentially cuts out character patterns included in image data read by the document reading unit 401 and based on a result of comparison with a pattern dictionary 403. Characters are recognized, and the obtained recognition result is used as a similar character discriminating device 21
0 to a document file system (not shown) or the like.

In the similar character discriminating device 210 shown in FIG. 3, the character string input section 211 corresponds to the character set input means 111 described in claim 2, and is a sequence corresponding to a predetermined range on the document. Of the character pattern and the recognition result from the pattern recognition unit 402,
12 and a configuration used for processing by the height distribution evaluation unit 213.

The character string input unit 211 operates as, for example, a character set input unit 111 which will be described later in the additional disclosure item 6, and sequentially inputs a series of character patterns and recognition results corresponding to each line on the document as a determination target. The configuration may be such that The character type discriminating unit 212 corresponds to the character type discriminating unit 114 described in claim 2, and each character belonging to the recognition result for one line received from the character string input unit 211 is shown in FIG. It is configured to determine to which of the character types it belongs and to provide the determination result to the processing of the determination control unit 214.

As shown in FIG. 4 (a), the similar character type for an alphabetic character is a type to which characters represented by a character pattern in which a set of uppercase and lowercase characters are similar to each other belong, for example, "C, O, S, U, V, W, X, Z"
And lowercase letters "c, o, s, u, v, w, x, z"
Contains. In addition, the uppercase type is a type to which a character that is in contact with both an uppercase line and a base line respectively given as an upper tangent and a lower tangent of an uppercase letter, and whose uppercase and lowercase letters are represented by different character patterns, For example, the uppercase letters "A, B, D, E, F, G,
H, K, L, M, N, R, T, Y "and lowercase letters" b, d, f, h, k ".

On the other hand, the lower case type is a type to which a character which is in contact with both a lower case line given as an upper tangent of a lower case and the above-mentioned reference line and whose upper case and lower case are represented by character patterns different from each other belongs. , Lowercase letters “a, e, m, n, r”. The middle character type is a type to which a character which is in contact with the above-mentioned reference line and whose upper end of the character pattern is located between a lower case line and an upper case line belongs to, and includes, for example, lowercase letters "i, t". .

On the other hand, a character that is in contact with an uppercase line and has a character pattern protruding below the reference line, such as an uppercase letter “Q”, belongs to the lowercase uppercase type and is represented by a lowercase letter “q, y”. In addition, characters that touch the lowercase line and have a character pattern protruding below the reference line belong to the lowercase type,
Characters whose upper end of the character pattern is halfway between the lowercase and uppercase lines and whose character pattern protrudes below the base line, such as the lowercase letter “g, j”, belong to the lower middle character type .

Note that the lowercase letter “p” and the uppercase letter “P” are of a special similar character type, and when determining whether a character belonging to this is uppercase or lowercase, the base line of the lowercase letter “p” is determined. The height of the upper part may be the height of the character. On the other hand, the uppercase letters “I, J” and the lowercase “l” are of the target exclusion type because the character type estimation system is low, and these characters are not used in the discrimination processing described later.

In FIG. 3, a height distribution evaluator 21 is provided.
3 obtains the character height and distribution range of each character included in the corresponding line based on the character pattern for one line received from the character string input unit 211, and calculates the obtained distribution range as the character range of each character. Discrimination control unit 2 together with height information indicating the height
14 is provided. The discrimination control unit 214 illustrated in FIG. 3 performs the discrimination analysis processing unit 215 based on the discrimination result of the character type discrimination unit 212 and the distribution range of the height of the character pattern obtained by the height distribution evaluation unit 213. , Controls the operations of the height determination processing unit 216 and the height ratio determination processing unit 217, and determines the similar characters included in the pattern recognition result received via the character string input unit 211 based on the determination results obtained by these units. Is determined and sent as a recognition result to a document file system (not shown).

FIG. 5 is a flowchart showing the similar character determining operation. When a pattern recognition result for one line is input by the character string input unit 211 shown in FIG.
2, the character type to which each character included in the pattern recognition result belongs is determined (steps 301 and 302).
If the similar character type is not included in the set of character types obtained in step 302, the determination control unit 214 proceeds to step 310 as a negative determination in step 303 and receives the same via the character string input unit 211. What is necessary is just to send the pattern recognition result as it is to the document file system and end the processing.

On the other hand, when the similar character type is included in the set of character types obtained in the above-described step 302, the discrimination control unit 214 refers to the evaluation result by the height distribution evaluating unit 213 and inputs the same. It is determined whether the height of the character pattern corresponding to the character string is uniform (step 30).
4). When the distribution range obtained by the height distribution evaluation unit 213 is equal to or more than the predetermined threshold V1, the discrimination control unit 214
It is determined that the input character string to be distinguished satisfies the conditions for applying the discriminant analysis, and as a negative determination in step 304, the height information obtained by the height distribution evaluation unit 213 is processed by the discriminant analysis processing unit 215. It is good to serve.

In response, the discriminant analysis processing section 215 performs discriminant analysis processing (step 305). FIG.
The figure explaining discriminant analysis is shown in FIG. Characters belonging to the similar character type can be classified into a class having a character height equivalent to uppercase letters and a class having a character height equivalent to lowercase letters.

On the other hand, if the character string to be discriminated contains a sufficient number of characters, the distribution of the height of each character contained in this character string is, as shown in FIG. It can be expressed as the sum of a normal distribution centered on the height and a normal distribution centered on the height of lowercase characters. Accordingly, the discriminant analysis processing unit 215 obtains a threshold k for separating the two normal distributions described above using a discriminant analysis technique, and based on the threshold k, a similar character type character included in the character string to be discriminated is determined. What is necessary is just to judge whether it is a capital letter or a small letter.

FIG. 6B shows a detailed configuration diagram of the discriminant analysis processing unit. In FIG. 6, the histogram creation unit 221 includes:
Based on the height information received from the discrimination control unit 214,
A histogram representing the distribution of the height of each character included in the character string to be determined is created and provided to the processing of the degree-of-separation calculating unit 223. On the other hand, the threshold determination unit 222 responds to the instruction from the optimization control unit 224 to set the threshold k
Is adjusted and provided to the process of the degree-of-separation calculating unit 223. The degree-of-separation calculating unit 223 converts the distribution represented by the above-described histogram based on the threshold value k received from the threshold value determining unit 222. In the case of separation, the configuration is such that a degree of separation λ indicating the sum of the probability of erroneous determination is obtained.

Further, the optimization control unit 224 controls the threshold value determination processing by the threshold value determination unit 222 according to the value of the above-mentioned separation degree λ, and compares the threshold value k when the separation degree λ becomes the minimum. The configuration is provided for the processing of the unit 225. The threshold k obtained in this manner corresponds to the distribution of character heights corresponding to characters having the same character height as uppercase letters shown in FIG. 4 shows the boundary between the character height distribution and the character height distribution.

Therefore, the comparison processing unit 225 receives the character height corresponding to the similar character type character to be discriminated from the histogram creation unit 221 and, based on the comparison result between the character height and the threshold k, determines the similar character. Is determined to be a capital letter or a small letter, and the determination result is determined by the determination control unit 214.
Return it to

In response, the discrimination control unit 214
In step 306, the recognition result for the similar character included in the recognition result by the pattern recognition unit 402 is determined based on the received determination result, and the obtained recognition result is sent to a document file system or the like to perform processing. It should just end. On the other hand, if the height of the character string for one line to be determined is uniform, an affirmative determination is made in step 304 described above, and the determination control unit 214 determines in step 30
In 7, it is determined whether or not the result of the determination by the character type determination unit 212 includes a character belonging to a character type serving as a reference for the character height of uppercase characters or the character height of lowercase characters.

In the case of an affirmative determination in step 307,
The discrimination control unit 214 operates as the reference character extracting means 131 described later in the additional disclosure item 1, and outputs the information indicating the height and the character type of the obtained reference character and the information indicating the height of the similar character to be discriminated. To the height determination processing unit 216, and accordingly, the height determination processing unit 216
A determination process based on the height of the reference character is performed (step 3).
08).

At this time, the discrimination control unit 214 preferentially selects a character belonging to an uppercase type or a lowercase type as a reference character. Select a character belonging to the middle or lower case type as the reference character, and add additional disclosure
May be provided for the processing of the height determination processing unit 216 corresponding to the height determination means 132 described in.

FIG. 7 is a flowchart showing the discrimination operation based on the character height. The height determination processing unit 216 determines in step 311
First, it is determined whether or not the reference character type received in step 311 is an uppercase type (step 312). In the case of an affirmative determination, the height determination processing unit 216 determines that the similar character to be determined is uppercase. Then (Step 313), the determination result is returned to the determination control unit 214 (Step 314), and the process may be terminated.

On the other hand, in the case of a negative determination in step 312, it is determined whether or not the received reference character type is a lower case type (step 315). In the case of an affirmative determination, the height determination processing unit 216 determines It is determined that the target similar character is a lowercase letter (step 316), and the determination result is determined by the determination control unit 21.
4 (step 314), and the process may be terminated. On the other hand, in the case of a negative determination in step 315, the height determination processing unit 216 compares the character height Hs of the reference character with the character height Hx of the similar character to be determined and determines that they are substantially equal (step 317). The process proceeds to step 313 to determine that the character is a capital letter, and if the determination is negative in step 317, the process proceeds to step 316 to determine that the character is a small letter.

Upon receiving the determination result obtained in this manner, the determination control unit 214 executes step 30 shown in FIG.
Then, the process may be terminated by reflecting the determination result to the pattern recognition result and proceeding to step 6. By the way, when the character string to be determined does not include a character belonging to a character type that can be a reference, the determination control unit 214 determines the character height of the similar character to be determined as a negative determination in step 307. The information is sent to the height ratio determination processing unit 217. In response, the height ratio determination processing unit 217 performs determination processing based on the ratio between the character height of the corresponding similar character and the standard uppercase character height. Is performed (step 309).

FIG. 8 shows a detailed configuration of the height ratio determination processing section 217. In the height ratio determination processing unit 217 shown in FIG. 8, the circumscribed rectangle extraction unit 231 operates as the reference character extraction unit 131 described later in the additional disclosure item 2, and outputs each line on the document via the character string input unit 211. The corresponding character patterns are sequentially received, and the circumscribed rectangles of the character patterns corresponding to the respective characters included in the character pattern for one line are extracted. Line estimation unit 232, capital letter line estimation unit 233
And a lower-case line estimating unit 234.

The circumscribed rectangle extracting section 231 sends, for example, a circumscribed rectangle corresponding to the character pattern determined to be uppercase to the reference line estimating section 232 and the uppercase line estimating section 233, and outputs the character pattern determined to be lowercase. The circumscribed rectangle corresponding to the reference line estimating unit 232 and the lowercase line estimating unit 234
Should be sent to The reference line estimating unit 232 corresponds to the reference line estimating unit 136 described in the additional disclosure item 3 later. For the set of the midpoints of the lower bottom sides of the circumscribed rectangles received from the circumscribed rectangle extraction unit 231, for example, A straight line approximation process using the least squares method or the like is performed to estimate a reference line commonly circumscribing these circumscribed rectangles, and the estimated reference line is used for processing by the uppercase size estimator 235 and the lowercase size estimator 236.

On the other hand, the upper-case line estimating unit 233 corresponds to the upper-case line estimating unit 135 described in the additional disclosure item 3 later. For example, a straight line approximation process using the least squares method is performed to estimate a capital letter line that circumscribes these circumscribed rectangles in common, and is used for processing by the capital letter size estimation unit 235.

The upper case size estimating unit 2 shown in FIG.
35 is equivalent to the character height calculating means 137 described later in the additional disclosure item 3, and calculates the standard capital letter height based on the distance between the received capital letter line and the reference line. In addition to the processing of the height ratio calculation unit 239, the obtained uppercase height is stored in the size storage unit 238a.

As described above, the above-described reference line estimating unit 232
By the operation of the capital letter line estimating unit 233, the reference line and the capital letter line that accurately reflect the size of the characters included in the line on the current document are precisely estimated. By providing the function, the function of the capital letter height estimating means 133 described later in the additional disclosure item 2 is realized, and based on the series of character patterns corresponding to each line received via the character string input unit 211, a standard capital letter An index indicating the character height can be dynamically calculated.

In FIG. 8, the selector 237a
Selects either the output of the uppercase size estimation unit 235 or the output of the size holding unit 238a in accordance with an instruction from the uppercase line estimation unit 233, and sets the uppercase character height Ha as the processing of the height ratio calculation unit 239. It is configured to be used for Here, the uppercase line estimating unit 233 instructs the selector 237a to select the output of the uppercase size estimating unit 235 when the uppercase line can be normally estimated, and when the uppercase line estimating process cannot be performed. The selector 237a may be instructed to select the output of the size holding unit 238a.

As described above, by controlling the selection operation of the selector 237a in accordance with whether or not the capital letter line has been correctly estimated, the character type serving as the reference for the character height of the capital letter in the range currently focused on. Regardless of whether or not there is a character belonging to, an appropriate index indicating the height of uppercase letters can be provided to the processing of the height ratio calculation unit 239. The height ratio calculation unit 239 receives the character height Hx of the similar character to be determined via the determination control unit 214, and calculates a height ratio Vh between the character height Hx and the above-described uppercase character height Ha. The height ratio determination unit 134 described later in the additional disclosure 2 is formed together with the threshold calculation unit 240 and the comparison processing unit 241 to be described later.

In FIG. 8, the comparison processing unit 241 compares the height ratio Vh with the threshold value Thv obtained by the threshold value calculation unit 240.
, And a determination result indicating whether the similar character to be determined is a capital letter or a small letter is returned to the determination control unit 214 in accordance with the obtained result. In FIG. 8, the threshold calculator 240 calculates a threshold Thv based on the ratio between the uppercase character height Ha described above and the lowercase character height Hb described later, and provides the threshold Thv to the processing of the comparison processor 241. And it is sufficient.

In the same manner as the above-described uppercase character height Ha, an index indicating the lowercase character height Hb is obtained, and the threshold value Thv for the height ratio can be dynamically determined. In FIG. 8, the lowercase line estimating unit 234 includes a circumscribed rectangle extracting unit 23.
For the set of the midpoints of the upper bottom sides of the circumscribed rectangles received from No. 1, the same processing as that of the above-described upper-case line estimation unit 233 is performed, the lower-case lines common to these circumscribed rectangles are circumscribed, and the lower-case size is estimated. The configuration is provided for the processing of the unit 236.

The lower case size estimating unit 236 obtains a standard lower case height based on the distance between the received lower case line and the reference line, and provides the same to the threshold calculating unit 240 via the selector 237b. The obtained lowercase height is stored in the size storage unit 238b. The selector 237b selects one of the output of the lowercase size estimating unit 236 and the output of the size holding unit 238b in response to an instruction from the lowercase line estimating unit 234, and sets the lowercase character height Hb as the lowercase character height Hb. What is necessary is just to set it as the structure provided for a process.

Thus, the character height Ha of the capital letter, which is a criterion for determination based on the height ratio, is determined according to the characteristics of each line on the document.
And a threshold value Thv for discrimination is dynamically determined, and similar characters can be discriminated in accordance with these criteria. Here, step 309 shown in FIG. 5 is executed because the character string to be discriminated has a uniform height and includes a character belonging to a character type serving as a standard for uppercase or lowercase height. This is a very rare case.

At this time, since the estimation process cannot be performed in both the capital letter line estimation unit 233 and the small letter line estimation unit 234, the capital letters held in the size holding units 238a and 238b by the selectors 237a and 237b, respectively. The height Ha and the lowercase character height Hb are selected, and the height ratio calculation unit 239, the threshold value calculation unit 240, and the comparison processing unit 241 perform determination processing based on the height ratio.

In this case, as described above, the uppercase character height and the ratio of uppercase and lowercase character heights in the line recognized earlier than the line on the current document of interest are used as references. By performing the determination process based on the character height, it is possible to determine whether the similar character to be determined is a capital letter or a small letter. The uppercase line estimator 233, the reference line estimator 232, the lowercase line estimator 234, the uppercase size estimator 235, and the lowercase size estimator 236 shown in FIG. Is accurate unless there is a change in characteristics such as font switching or point number switching between the line and the current line of interest.

It is also possible to determine similar characters based on the result of comparing the height ratio obtained by the height ratio calculation unit 239 with a fixed threshold. In this case, the height ratio determination processing unit 217 is configured by excluding the lowercase line estimation unit 234, the lowercase size estimation unit 236, the selector 237b, the size holding unit 238, and the threshold value calculation unit 240 illustrated in FIG. A fixed threshold determined based on the ratio of the character height of uppercase characters to the character height of lowercase characters in a simple font may be provided to the processing of the comparison processing unit 241.

Here, as described above, the character string input unit 2
In the case where the similar character discriminating process is performed for each line on the document by the operation of 11, the discriminant analysis processing unit 215, the height determination processing unit 216, and the height ratio determination processing unit 217
A determination index for determining whether a similar character is uppercase or lowercase is obtained for each line on the document.

Therefore, for example, as in a footnote,
Even when a document contains a document with a portion where the number of points changes for each line, it is possible to dynamically change the discrimination index for each line and accurately identify similar characters included in each line. is there. Further, each unit constituting the similar character discriminating apparatus shown in FIG. 3 can be realized by software, respectively.
The computer can execute the character set input procedure, the height distribution evaluation procedure, the discriminant analysis processing procedure, the classification procedure, and the individual determination procedure described in, and such a program is stored in a storage medium such as a floppy disk or a CD-ROM. By recording and distributing, accurate similar character discrimination processing according to the present invention can be provided to a wide range of users.

Next, a description will be given of a method of performing similar character discrimination processing by mutually using Japanese characters and English characters.
Here, as shown in FIG. 9A, a standard hiragana capital letter is often represented by a character pattern that is slightly larger than a standard English capital letter. Therefore, for example, when mainly recognizing a document described in English characters, FIG.
As shown in (b), the Japanese character discriminating unit 251 provided in the character string input unit 211 discriminates a character serving as a reference for the height hn of the hiragana capital letter, and the normalization processing unit 252
The height of the character pattern representing these characters may be multiplied by an appropriate normalization constant Cs to be provided to the processing of the character type determination unit 212 and the height distribution evaluation unit 213.

The Japanese character discriminating section 251 corresponds to the language discriminating means 121 described later in the additional disclosure item 4. Based on the pattern recognition result received via the character string input section 211, for example, FIG. A configuration may be adopted in which the hiragana characters “ka, ki, ku, ke, sa, shi, su, so” shown in b) are determined as reference characters. The normalization processing unit 252 corresponds to the normalization unit 122 described later in the additional disclosure item 4. For example, a standard hiragana capital letter height hn and a standard alphabetical capital letter height hn
The normalization constant Cs is obtained in advance based on the ratio to he, and the character height hx corresponding to each character determined by the Japanese character determination unit 251 is normalized by the above-described normalization constant Cs, and Along with the recognition result, the character type discriminating unit 2
12 and the character height distribution evaluation unit 213.

Conversely, when Japanese characters are mainly recognized, if the height of English characters mixed in a document described in Japanese characters is normalized to the height of Hiragana capital letters, The similar character "A,
I, u, ue, oh, tsu, ya, yu, yo, ぁ, ぃ, ぅ, ぇ,
ぉ, ゃ, ゃ, ゅ, 」”.

On the other hand, the character string input unit 211 operates as a character set input unit 111 described later in the additional disclosure item 5, and determines, for example, a pattern recognition result for a plurality of lines within a reading range by the document reading unit 401. When the character string is input as a character string, the discriminant analysis processing unit 215
The height determination processing unit 216 and the height ratio determination processing unit 217 can determine a determination index for determining whether a similar character is uppercase or lowercase for each of the reading ranges.

In this case, since a sufficient number of samples can be secured, the discriminant processing by the discriminant analysis processing unit 215 can be performed reliably. According to the discriminant analysis processing based on sufficient samples, an appropriate Since it is possible to estimate the threshold value k with sufficient accuracy, it absorbs subtle changes in the size of characters due to changes in the emphasis expressions and fonts included in the area to be determined, and the above-described character string to be determined Can be reliably distinguished between uppercase and lowercase.

On the other hand, when the area to be determined is enlarged in this way, it is not possible to sufficiently cope with a local change in the number of points or the like. Conversely, the character string input unit 211
Operates as the character set input unit 111 described later in the additional disclosure item 7, and inputs the pattern recognition result in units of words. In the case where the discrimination analysis processing unit 215, the height determination processing unit 216, and the height ratio determination The processing unit 217 can dynamically determine, for each word, a determination index for determining whether a similar character is a capital letter or a small letter.

In this case, since the number of samples is extremely small, the estimation accuracy of the discrimination index obtained by the discriminant analysis processing unit 215, the height determination processing unit 216, and the height ratio determination processing unit 217 may decrease. There is. On the other hand, as described above, based on the discrimination index dynamically obtained for each word, the similar characters included in the word are discriminated to include, for example, a portion where the number of points changes in word units. Even when the document is a manuscript, similar characters can be determined.

[0059]

As described above, according to the first, second, and tenth aspects of the present invention, based on the feature relating to the height of a character pattern included in a predetermined range read from a document. By performing the discriminant analysis process or the individual judgment process, similar characters within the predetermined range can be determined based on the discrimination index dynamically obtained for the predetermined range. It is possible to accurately determine similar characters by absorbing changes in character patterns.

With respect to the above description, the following items are further disclosed. Additional Disclosure 1. 3. The similar character discriminating apparatus according to claim 2, wherein the individual judging unit 115 receives the classification result by the classifying unit 114, and converts a character pattern corresponding to a character classified into a character type that can be a reference for the height of uppercase letters or lowercase letters. A reference character extracting unit 131 that extracts a reference character pattern, and a height that determines a similar character recognition result according to a comparison result between the height of the reference character pattern and the character pattern corresponding to the similar character to be determined. Judgment means 13
2 is provided.

In the similar character discriminating apparatus according to additional disclosure 1, the height determining means 132 directly compares the height of the reference character pattern extracted by the reference character extracting means 131 with the height of the character pattern of the similar character. Accordingly, similar characters can be accurately determined based on the height of the reference characters included in the character set. As described above, when the additional disclosure item 1 is applied, the difference between fonts can be improved by using the height of a character whose recognition result has been determined by the pattern recognition means as a determination index for determining uppercase and lowercase. Regardless of a change in the character pattern due to the emphasized expression, similar characters can be accurately determined.

Additional Disclosure Item 2. 3. The similar character discriminating apparatus according to claim 2, wherein the individual judging means 115 comprises
A reference character extraction unit 131 that receives a classification result according to 14 and extracts a character pattern corresponding to a character classified into a character type that can be a reference for the height of uppercase or lowercase letters as a reference character pattern;
Uppercase height estimation means 1 for estimating standard uppercase height
33, a height ratio determination for determining a recognition result of similar characters based on a ratio of a capital letter height obtained by the capital letter height estimating means 133 to a character pattern corresponding to a similar character to be determined. And means 134.

In the similar character discriminating apparatus according to Additional Disclosure 2, the capital letter height estimating means 133 estimates the standard capital letter height based on the height of the reference character extracted by the reference character extracting means 131. The height ratio determination means 13
According to FIG. 4, the similar character can be determined based on the ratio between the standard uppercase letter height and the height of the similar character to be determined.

Additional Disclosure Item 3. 5. The similar character discriminating apparatus according to claim 4, wherein the capital letter height estimating means 133 is a set of middle points of the upper bottom of a circumscribed rectangle corresponding to a reference character pattern classified into a character type that can be a reference for the capital letter height. Capital line estimating means 135 for estimating a capital line circumscribing the upper part of a character pattern representing a standard capital letter based on
Based on the set of midpoints on the lower base of the circumscribed rectangle corresponding to the reference character pattern categorized as a character type that can be the standard for uppercase or lowercase characters, a character pattern representing standard uppercase and lowercase characters It is provided with a reference line estimating means 136 for estimating a reference line whose lower part is circumscribed, and a character height calculating means 137 for determining the distance between the capital letter line and the reference line and outputting the distance as a standard capital letter height. It is characterized by the following.

The similar character discriminating apparatus according to additional disclosure item 3 accurately estimates a capital letter line and a reference line based on the heights of a plurality of reference characters by the operation of the capital letter line estimation means 135 and the reference line estimation means 136. Therefore, the character height calculator 137 can accurately determine the height of a standard uppercase letter.

Therefore, if the additional disclosure item 2 and the additional disclosure item 3 are applied, the standard capital letter height can be precisely determined based on the character pattern corresponding to the character whose recognition result has been determined by the pattern recognition means. By estimating and using the ratio to the standard uppercase letter height as a discrimination index, similar characters can be accurately discriminated irrespective of font differences or changes in character patterns due to emphasized expressions.

Additional Disclosure Item 4. In the similar character discriminating apparatus according to claim 2, the character set input means 111 is mainly composed of a language discriminating means 121 for discriminating a language to which each character included in the character string received from the pattern recognizing means 101 belongs, and A configuration including a normalizing means for normalizing the size of the corresponding character pattern to each character determined as a foreign language character belonging to a language other than the main language to be handled in accordance with the size of the character belonging to the main language; It is characterized by being.

In the similar character discriminating apparatus according to the additional disclosure item 4, each character determined as a foreign language character by the language discriminating means 121 is included in the character set by the normalizing means 122 performing normalization processing. Foreign language characters can be used to determine similar characters. This additional disclosure item 4
, By normalizing the size of foreign language characters based on the standard capitalization of main language characters,
Based on a character set that includes foreign language characters as main language characters,
Since it is possible to obtain the discrimination index, it is possible to improve the accuracy of discriminating similar characters in a document in which characters used in a plurality of languages are mixed.

Additional Disclosure Item 5. 3. The similar character discriminating apparatus according to claim 2, wherein the character set input unit 111 obtains, by the pattern recognition unit 101, a character pattern included in a predetermined reading area including the similar character to be discriminated in the original document and these character patterns. The received recognition result is input.

The similar character discriminating apparatus according to the additional disclosure item 5 operates the character set input means 111 to input a character pattern included in the reading area and its recognition result as a character set, thereby providing the discriminant analysis processing means 113 and the individual The determination means 115 can dynamically determine a determination index for determining whether a similar character is a capital letter or a small letter for each of the above-described reading regions.

When the additional disclosure item 5 is applied and the discrimination index is obtained for each reading area, similar characters can be discriminated accurately by the discriminant analysis processing based on sufficient samples. Additional Disclosure Item 6. 3. The similar character discriminating apparatus according to claim 2, wherein the character set input unit 111 includes a character pattern corresponding to one line including a similar character to be determined in the document and a pattern recognition unit 101 for these character patterns.
And a recognition result obtained by the above.

The similar character discriminating apparatus according to the additional disclosure item 6 inputs a character set corresponding to one line on a document by the operation of the character set inputting means 111, so that the discrimination analysis processing means 113 and the individual discriminating means 115 The determination index for determining whether the similar character is uppercase or lowercase can be dynamically obtained for each line on the document and can be determined.

Additional Disclosure Item 7. 3. The similar character discriminating apparatus according to claim 2, wherein the character set input unit 111 includes a character pattern corresponding to a word including a similar character to be determined in the document and a recognition obtained by the pattern recognizing unit 101 for these character patterns. And a result input unit. In the similar character discriminating apparatus according to the additional disclosure item 7, when the character set corresponding to the word is input by the operation of the character set input unit 111, the similar character is capitalized in the discriminant analysis processing unit 113 and the individual determination unit 115. It is possible to dynamically determine and determine a determination index for determining whether a word is a lowercase or a lowercase for each word.

When the additional disclosure item 6 or 7 is applied to determine the discrimination index for each line or word on the manuscript, it is possible to flexibly follow a local change in the number of points, etc. Can be determined.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating the principle of a similar character determination method according to the present invention.

FIG. 2 is a principle block diagram of a similar character discriminating apparatus of the present invention.

FIG. 3 is a diagram showing a configuration of a character recognition device to which the similar character discrimination device of the present invention is applied.

FIG. 4 is a diagram illustrating an example of a character type.

FIG. 5 is a flowchart illustrating a similar character determining operation.

FIG. 6 is a diagram illustrating discriminant analysis.

FIG. 7 is a flowchart illustrating a determination operation based on a character height.

FIG. 8 is a detailed configuration diagram of a height ratio determination processing unit.

FIG. 9 is a diagram illustrating characteristics of Japanese characters and English characters.

FIG. 10 is a configuration diagram of a character recognition device to which the invention technique described in Japanese Patent Application Laid-Open No. 3-111983 is applied.

FIG. 11 is a diagram illustrating a conventional determination method based on a blank ratio.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 101 Pattern recognition means 111 Character set input means 112 Height distribution judgment means 113 Discrimination analysis processing means 114 Classification means 115 Individual judgment means 121 Language discrimination means 122 Normalization means 131 Reference character extraction means 132 Height judgment means 133 Uppercase letter height estimation Means 134 Height ratio determining means 135 Uppercase line estimating means 136 Reference line estimating means 137 Character height calculating means 210 Similar character discriminating device 211 Character string input unit 212 Character type discriminating unit 213 Height distribution evaluating unit 214 Discrimination control unit 215 Discrimination Analysis processing unit 216 Height determination processing unit 217 Height ratio determination processing unit 221 Histogram creation unit 222 Threshold determination unit 223 Separation degree calculation unit 224 Optimization control units 225, 241 Comparison processing unit 231 Circumscribed rectangle extraction unit 232 Reference line estimation unit 233 Upper case line estimation unit 234 Small Character line estimation unit 235 Uppercase size estimation unit 236 Lowercase size estimation unit 237 Selector 238 Size holding unit 239 Height ratio calculation unit 240 Threshold calculation unit 401 Document reading unit 402 Pattern recognition unit 403 Pattern dictionary 404 Similar character extraction unit 405 Output processing unit 410 Similar character determination unit 411 Blank ratio calculation unit 412 Comparator

Claims (3)

[Claims] 1. A character recognition device comprising a pattern recognition means for recognizing each character written on a document based on a character pattern read from the document, wherein uppercase and lowercase characters are represented by similar character patterns. Receiving a character pattern string corresponding to a character string in a predetermined range including similar characters, and a recognition result performed on the character pattern string by the pattern recognition unit, the height of the character pattern included in the character pattern string Is determined whether or not is uniform, if the heights of the character patterns included in the character pattern sequence varies, by discriminant analysis of the height of the character patterns included in the character pattern sequence, capital letters Classified into uppercase class with height equivalent to lowercase and lowercase class with height equivalent to lowercase. The recognition result of the similar character is determined based on the classification result. If the height of the character pattern included in the character pattern sequence is uniform, the character pattern is determined for the determined character included in the recognition result. Are classified into a plurality of character types having different distribution characteristics in the height direction, and the height of a character pattern corresponding to the determined character classified into the plurality of character types and the height of a character pattern corresponding to the similar character. Determining a recognition result corresponding to the similar character according to a result of comparison with the similar character. 2. A character recognition device comprising a pattern recognition means for recognizing each character written on a document based on a character pattern read from the document, wherein uppercase and lowercase characters are represented by similar character patterns. Character set input means for receiving and inputting a character pattern string corresponding to a character string in a predetermined range including similar characters, and a recognition result performed on the character pattern string by the pattern recognition means; and Height distribution determining means for determining whether or not the height of each character pattern included in the character pattern is uniform; and the character pattern string according to a determination result that the height of each of the character patterns varies. Based on the discriminant analysis of the height of the character pattern included in the uppercase, the uppercase class with the height equivalent to the uppercase and the small sentence with the height equivalent to the lowercase Character class, and based on this classification result,
Discriminant analysis processing means for determining a recognition result of the similar character to be determined; and for a determined character included in the recognition result, according to a determination result that the height of each of the character patterns is uniform. A classification means for classifying into a plurality of character types having different distribution characteristics in the height direction of the character pattern; and a height of the character pattern corresponding to the similar character, and a determined character classified into the plurality of character types. A similar character discriminating device, comprising: individual determining means for individually determining a recognition result corresponding to each of the similar characters based on a height of a corresponding character pattern. 3. A character recognition device comprising a pattern recognition means for recognizing each character written on a document based on a character pattern read from the document, wherein uppercase and lowercase characters are represented by similar character patterns. A character set sequence for receiving and inputting a character pattern string corresponding to a character string in a predetermined range including similar characters, and a recognition result performed on the character pattern string by the pattern recognition means; and A height distribution determining procedure for determining whether the height of each character pattern included in the character pattern is uniform, and the character pattern string according to a determination result that the height of each character pattern varies. Based on the discriminant analysis of the height of the character pattern included in the uppercase, the uppercase class with the height equivalent to the uppercase and the small sentence with the height equivalent to the lowercase Character class, and based on this classification result,
A discriminant analysis processing procedure for determining a recognition result of the similar character to be determined, and a determined character included in the recognition result according to a determination result that the height of each character pattern is uniform. A classification procedure of classifying into a plurality of character types having different distribution characteristics in the height direction of the character pattern; and a height of a character pattern corresponding to the similar character, and a determined character classified into the plurality of character types. A storage medium storing a similar character determination program for causing a computer to execute an individual determination procedure for individually determining a recognition result corresponding to each of the similar characters based on a height of a corresponding character pattern.